Despite the ubiquity of delusional information processing in psychopathology and everyday life, formal characterizations of such inferences are lacking. In this article, we propose a generative framework that entails a computational mechanism which, when implemented in a virtual agent and given new information, generates belief updates (i.e., inferences about the hidden causes of the information) that resemble those seen in individuals with delusions. We introduce a particular form of Dirichlet process mixture model with a sampling-based Bayesian inference algorithm. This procedure, depending on the setting of a single parameter, preferentially generates highly precise (i.e. over-fitting) explanations, which are compartmentalized and thus can co-exist despite being inconsistent with each other. Especially in ambiguous situations, this can provide the seed for delusional ideation. Further, we show by simulation how the excessive generation of such over-precise explanations leads to new information being integrated in a way that does not lead to a revision of established beliefs. In all configurations, whether delusional or not, the inference generated by our algorithm corresponds to Bayesian inference. Furthermore, the algorithm is fully compatible with hierarchical predictive coding. By virtue of these properties, the proposed model provides a basis for the empirical study and a step toward the characterization of the aberrant inferential processes underlying delusions.
